2012 Cambridge Business & Economics Conference ISBN : 9780974211428 A Simulation Model To Forecast Future Cash-Flows – A Financial Risk Management Tool Ralf Bernsau Karlsruhe Institute of Technology (KIT) Institute of Applied Informatics and Formal Description Methods (AIFB) Karlsruhe, Germany +49 176 32189761 Ralf.Bernsau@student.kit.edu Andreas Vogel Karlsruhe Institute of Technology (KIT) Institute of Applied Informatics and Formal Description Methods (AIFB) Karlsruhe, Germany +49 721 60845393 andreas.vogel@kit.edu Detlef Seese Karlsruhe Institute of Technology (KIT) Institute of Applied Informatics and Formal Description Methods (AIFB) Karlsruhe, Germany +49 721 60846037 detlef.seese@kit.edu June 27-28, 2012 Cambridge, UK 1 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 A Simulation Model To Forecast Future Cash-Flows – A Financial Risk Management Tool ABSTRACT This article describes a simulation model which enables the user to forecast the possible trend of the companies’ Cash-Flow based on historical data, individual estimations, multivariate regression and the Value-at-Risk concept. The simulation model is able to simulate Cash-Flows for different individual scenarios and serves due to that as a financial risk management tool. One interesting feature is that the model uses not just time series of the relevant market parameters, the multivariate regression includes an additional extern parameter. This parameter describes the influence of the environment on the future price trends of the relevant market parameters. The implementation of the extern parameter follows a random walk. Therefore, the model is not just focusing on a small number of main market parameters, it also captures the development around these main parameters in one single factor. INTRODUCTION The competition conditions for the manufacturing industry have changed considerably in recent years. Due to the increasing globalization and the simultaneous fluctuations in international financial markets, companies face new challenges. As a result of the stronger integration of the economy and the consequent increase in volatility of commodity prices, equities, interest and exchange rates, the financial risks of companies have increased. Especially the recent past confirms that extreme market fluctuations occur at shorter time intervals and as a consequence the financial position and stability of banks, industrial and June 27-28, 2012 Cambridge, UK 2 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 commercial companies and even of nations is constantly challenged. The financial risk management is therefore an operational function which importance is increasing more and more. The success or failure of companies in a challenging environment is essential for their existence and global competitiveness. Established on these facts, a lot of companies and academics, such as the National Economic Research Association (NERA) with the cooperation of the Harvard professor Jeremy Stein (2000) or the RiskMetrics Group by Alvin J. Lee (1999), developed several Financial Risk Management Tools to simulate future price trends of stocks, commodities, interest and exchange rates and so forth. However, the opinions and approaches of these academics and professionals differ. That’s why Jan Duch (2006) subdivides the different approaches in two groups. One group is called the Top-Down approach and the other group is called the Bottom-Up approach. According to Jan Duch (2006) the aim of the Bottom-Up approach is to make a statement about the probability that a certain future Cash-Flow adopts a specific value due to the influencing factors. Attributed to the required knowledge of the business-related effect relationships, the approaches are called internal models. Additionally, the approach is closely based on the Value-at-Risk concept. Therefore, it is necessary to implement these models to start with searching and identifying market-price-based risk factors, which have a significant impact on the Cash-Flow. Ongoing, the identified financial risks are analyzed by using the exposure maps according to their importance for the outcome. After that the identified risks are brought into a functional relationship to simulate the future Cash-Flows. Finally, the calculation of the Cash-Flow-at-Risk is realized with a user-specific confidence level. On one hand, Chris Turner (1996) emphasized that the Bottom-Up approach is simple in the way of the intuitive interpretation of the result. It returns a value, which is exceeded with a given probability. In addition, it is possible to specify the risk of the probability of deviation from an June 27-28, 2012 Cambridge, UK 3 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 expected value or a quasi-reliable Cash-Flow. On the other hand, Turner states that the Bottom-Up models can be very complex. The calculation depends very much on the interplay of the various influencing factors. It is necessary to compute correlations between the individual parameters, which means that a big data base of time series has to be available. Furthermore, according to Turner (1996) you have to take in consideration how well the identified risk factors explain the Cash-Flow. A couple of various different methods are available to forecast future market prices. One method, which is analyzed by J. Kim, A. Malz and J. Mina (1999), is based on implied volatilities for short time periods by using deterministic forward prices. Moreover, according to Lee (1999) a simulation with a random walk can be implemented, based on historical moments of the distribution of the influencing factors. Finally, Lee (1999) of the RiskMetrics Group presented an estimation of the CashFlow development by using econometric methods with a so-called Vector Error Correction Model (VECM). The perspective of the Bottom-Up approach is not undisputed, because of the many interdependencies in terms of complexity. Stein (2000) emphasized that the danger might be large, to observe measurable risks, but easily to ignore other non-financial risks. Unlike the Bottom-Up approaches, the Top-Down approaches don’t consider separate individual risk factors. The first Top-Down approaches are the regression models and introduced by Bartram (1999). The regression models put the Cash-Flow in direct focus. Thus a study of individual or even entire risk exposures is also possible for non-company employees. The basis is the use of exclusively public capital market data. Therefore, based on Bartram the regression models are also called external regression models. The estimation of the volatility of the Cash-Flows is based on historical deviations. Through this procedure all risks, not only financial, also operational influences are taken into account, mentioned Stein (2000). Internal Cash-Flow data is either not available or it is in insufficient amount. Therefore, to explain the variation, June 27-28, 2012 Cambridge, UK 4 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 the models resort to regression analysis for example, with the stock returns or various capital market data. Another Top-Down model is the benchmark model, which estimates on historical Cash-Flow distributions and even to some extents from Cash-Flow distribution of competing companies, the variations. This benchmark model was developed by U.S. companies. The goal of this approach is to picture a company-wide aggregate risk, without using individual market parameters as the previous models. Simply put, it compares historical Cash-Flows from other companies in the same sectors with the company looked at, and then draws conclusions about possible future Cash-Flow trends. Therefore, no detailed knowledge is required about internal relationships in order to make a statement about the risk exposure of the company. The consulting firm National Economic Research Associates (NERA) from New York developed in 2000 such a model called C-FAR, where C-FAR stands for CashFlow-at-Risk. Like the two previous simulation models this model is also based on historical Cash-Flow time series. Although Stein (2000) highlighted in his paper, that the general problem is, that there are not enough empirical Cash-Flows existing, especially due to changes in corporate structure or company size. The model, which will be explained in this article, belongs to the Bottom-Up approach. The aim of that developed and implemented model is to anticipate future market movements and to measure the hazard on company’s financial strength and stability. The model is able to consider many single factors, which directly or indirectly influence the company’s wealth and economic strength. As a model stays always a model and can never predict the future with absolute certainty, nevertheless it is fundamental for global and international as well as national active companies to manage their risks. The model gives a good indication on possible future situations and simulates different economical scenarios. The company using the modeli will be able to deal with their risk and to get a better overview about coherences, June 27-28, 2012 Cambridge, UK 5 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 market factors and the consequences of their movements to evaluate the simulated world against reality. The model uses regression analysis and historical time series. However, none of the outlined models are able to describe the development of the environment and include the trend in the prediction of future prices. The article starts with explanations and descriptions of our simulation model and will continue with providing analyses and the evaluation of the model. The summary will conclude the described explanations and analyses of our simulation model. THE MODEL Portfolio optimization is primarily understood as the ability to simulate and evaluate a combination of several products. For economic reasons, it is realistic that a company produces and sells several products. Every company is able in our model to produce multiple products with different features. On one hand, the company has to determine for each product in its portfolio, the income elasticity and price elasticityii. On the other hand, a product-based allocation of sales, the expected value and standard deviation of the sales need to be indicated. Furthermore, each product is assigned a price, which is ideally above the production costs in order to realize a profit margin. Finally, each product must be assigned a breakdown in percentage proportion of the raw materials required. Through that, different products can be simulated with different features. Through the implementation of the individual products, the total sales of a company results from the sum of the sales of the individual products. The single turnovers arise from the sales of the products in Germany. The price of each product develops analogous to the June 27-28, 2012 Cambridge, UK 6 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Consumer Price Index (CPI) and automatically adjusts to the corresponding demand. Thus, the formula to show the price trend is: πππππ’ππ‘ ππππππ₯,π‘ = πππππ’ππ‘ ππππππ₯,π‘−1 ∗ πΆππΌπ‘ πΆππΌπ‘−1 (1) with π₯: πππππ’ππ‘ The turnover itself is derived from the sales of the previous period multiplied by the domestically economic development, the gross domestic product (GDP), is due to the income elasticity of demand. In addition, the turnover is influenced by the pricing of the product and weighted by the price elasticity. Alike the model considers diverse volatilities in sales, which can occur due to production, demand or external reasons. Fluctuations in production may arise for example, through line stoppage, staff absences or raw material supplies. This fluctuation is realized through the generation of a normally distributed random variable using the polar method of George Marsagliaiii. Thus, the calculation of product-specific sales results from the formula: πππππ π₯,π‘ = πππππ π₯,π‘−1 ∗ [1 + ( πΊπ·ππ‘ − 1) ∗ ππ₯,π ] πΊπ·ππ‘−1 (2) πΆππΌπ‘ − 1) ∗ ππ₯,π ] πΆππΌπ‘−1 ∗ [1 + π πππππ ππππππππ(π, π)] ∗ [1 + ( with π₯: ππ₯,π : ππ₯,π : π: π: πππππ’ππ‘ πΌπππππ ππππ π‘ππππ‘π¦ ππ π‘βπ πππππ’ππ‘ π₯ πππππ ππππ π‘ππππ‘π¦ ππ π‘βπ πππππ’ππ‘ π₯ πΈπ₯ππππ‘ππ π£πππ’π ππ π‘βπ π ππππ ππ‘ππππππ πππ£πππ‘πππ ππ π‘βπ π ππππ The company also incurred product-specific costs, which are explained in terms of material costs. As already mentioned, each commodity is traded in US-Dollar. Furthermore, June 27-28, 2012 Cambridge, UK 7 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 we make the assumption that the storage is refilled with raw materials at the beginning of each quarter to keep the storage constant. It will be bought just as much material as is required in order to realize the simulated sales. Thus, the product-specific material costs result from the sum of the portions of each commodity multiplied by its price and then adjusted for the πΈππ πππ· - exchange rate. πππ‘πππππ πΆππ π‘π πππ ππππ‘π₯,π‘ πππ· ∑ππ=1 πΉππππ‘ππππ ∗ ππππππ,π‘ ( ππππ ) = πΈππ πΈπ₯πβππππ π ππ‘ππ‘ (πππ· ) (3) with π₯: πππππ’ππ‘ π: ππππππ‘ πππππππ‘ππ (πΆπππππππ‘πππ ) The three above derived formulas are taken together and described as a so-called exposure map. This exposure map is individually constructible for each product of a company. For consideration of the overall risk, the risk potential of the various influencing factors or market parameters on the Cash-Flows needs to be identified. To use the Cash-Flowat-Risk it is necessary to determine the sensitivity to changes in the considered market parameters and for these circumstances the exposure map is used. According to Duch (2006) the exposure map is an economic mapping, which derives its focus on changes for the company's profit, due to changes in the revenue . Thus, the Cash-Flow for a product results by using the formula: πΆπΉπ₯,π‘ = (πππππ’ππ‘ ππππππ₯,π‘ − πππ‘πππππ πΆππ π‘π π₯,π‘ ) ∗ πππππ π₯,π‘ (4) with π₯: πππππ’ππ‘ Through the consideration of multiple products it is necessary, to calculate the total turnover of the company in a quarter, to add up the individual product-specific Cash-Flows. June 27-28, 2012 Cambridge, UK 8 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 π₯ (5) ππΆπΉπ‘ = ∑ πΆπΉπ₯,π‘ π₯=1 with πΆπΉπ₯,π‘ : πΆππ β − πΉπππ€ πππ πππππ’ππ‘ π₯ The five mentioned formulas are the base to calculate the Cash-Flows of a company. As outlined in the preparation of the exposure maps, we identified four key market parametersiv, which are essential to simulate the future Cash-Flows of an industrial enterprise. There is the gross domestic product (GDP) of the Federal Republic of Germany, the consumer price index (CPI), the long-term interest rate EURIBOR and the πΈππ πππ· - exchange rate. The gross domestic product (GDP) reflects directly the added value of the observed economy in the corresponding quarter. However, the problem with this measure is the frequency of data collection. Therefore to achieve a high statistical reliability, it is necessary to use a relatively long observation period of the past. Nevertheless, the quarterly GDP represents the key indicator for the quantification of the economy. In addition, we use seasonally and calendar adjusted values of the GDP, which the Federal Statistical Office of Germanyv makes available to avoid distortions of the actual economic development by seasonal influencesvi . By stating to the quarterly values in determining the economy, all other parameters need to be stated to quarterly values as well. This raises the fundamental question, whether we rely on average values of the quarters, or on a daily rate during the quarter. According to Siebert (2010) a calculation of quarterly averages distorts the actual volatility of the price developments, that’s why we use the closing price of the last trading day of each quarter. The next key indicator is the exchange rate. Based on the quarterly data supply of the GDP, the exchange rate will also be evaluated at the end of every quarter. The euro reference June 27-28, 2012 Cambridge, UK 9 2012 Cambridge Business & Economics Conference rates of the European Central Bank (ECB) deliver the needed data of the ISBN : 9780974211428 πΈππ πππ· - exchange rate. These euro reference rates are determined and published each business day through the participation of the European Central Bank and the National Central Banks and reflect the market price of the euro against major international currencies, stated Siebert (2010). The data series for the euro reference rates are accessible on the website of the German National Bankvii . It is also for the interest rate necessary to find a reference price, which reflects the general interest rate trends. Here arises the problem that, unlike the exchange rate many differing interest rates, for which banks lend money, are available. An appropriate index for the interest rate development is the Euro Interbank Offered Rate (EURIBOR). The EURIBOR is a reference rate, calculated by the ECB for time deposits in the interbank marketviii . This rate refers, in contrast to existing competition interest rates, such as the London Interbank Offered Rate (LIBOR), exclusively on the Euroix . Since the EURIBOR is calculated at different maturities and serves as a reference rate for floating rate notes and swaps, its use provides the representation for our interest rate. To map the corresponding long-term interest rates we chose the EURIBOR with the longest duration, twelve months. The historic EURIBOR time series are available on the website of the German National Bank. Like the GDP we use for the interest rate the quarterly value of the daily closing price of the last trading day of each quarter. As part of this market model we made the assumption that changes in commodity prices develop in line with prices in the economy. Against this background, the consumer price index (CPI), which is determined monthly by the Federal Statistical Office of Germany, is an accurate measure of consumer prices based on the Laspeyres price indexx. The CPI is used as a benchmark in wage negotiations and is constituted as the central indicator for the assessment of monetary developments in Germany. Furthermore, the changes in the consumer June 27-28, 2012 Cambridge, UK 10 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 price index or its internationally adapted form, called the harmonized consumer price index, are a measure of inflation in Germany. As we saw with the GDP, is it also necessary for the CPI to use seasonally and calendar adjusted values to avoid distortions of the actual development. Therefore, like Siebert (2010) we also used the value of the CPI of the last month of each quarter during the considered time period. Once the metrics are defined to quantify the key market parameters, the influences among the parameters need to be estimated. This turns out to be difficult, because on the theoretically profound basis you can find out, which factors affect another factor. The problem is that you cannot make precise statements about the strength and delay of the influence. Like the conventional risk models, the forecasts of the market parameters are calculated from returns. According to that, the influence of a parameter on another parameter is not based on the absolute value but on the return. The relevant relations of the derived market parameters were determined by using the theoretical principles. In addition, we used the t-statistics and the analysis of the correlations based on time series, for supporting the given theoretical relations. The correlation of two time series measures the significance of the direct influence of the return of a market parameter in t-1 on the return of a market parameter in t. The t-statistics measures the goodness of the gradient of the correlation and therefore rejects or supports the observed theoretical principles and estimated correlations of the market parameters. The news service Bloomberg provides the needed time series of the various market parameters. Beyond that, we used time series from the first quarter of 1990 to the second quarter of 2011, to provide a solid base of data to estimate influences and correlations between the market parameters. For example, we assessed a direct influence of the exchange rate on the long-term interest rate. The theory implies that expectations about the future exchange rate have a direct impact on net foreign investments. These net foreign investments arise from the difference between investments of residents abroad and of domestic investments of foreigners. In June 27-28, 2012 Cambridge, UK 11 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 absence of arbitrage the expectation of a re-valuation of the domestic currency leads to a higher demand for domestic bonds. Through that context, the domestic interest rate falls and the price of bonds rises. On the other hand, it is an expectation of a devaluation of domestic currency. As a result, the investment in foreign money is attractive. This leads to an outflow of capital to foreign countries. The reason for this observation is, that foreigners are now investing in their own country and nationals in the foreign country, based on the higher expected returns in the foreign country. Therefore, the net foreign investments decrease and the demand of domestic bonds regresses, whereby the interest rate raisesxi. As a result of this a connection between the development of the exchange rate and the development of the interest rate is supposed. Additionally, the estimated correlation of the time series of the exchange rate and the interest rate is supported by the t-statistics and therefore the estimated direct influence cannot be statistically rejected. We determined direct influences between the four key indicators and moreover a relationship to the individual raw materials. The direct influences between all the relevant market parameters are shown in table 1. The letter X implies a direct influence. The raw materials we used to simulate different products in our fictive simulation are aluminium (Alu.), copper, nickel and zinc. The times series of the four mentioned raw materials are also provided by Bloomberg and are based on the time period of 1990 to 2011. Once we evaluated all the relevant effect relationships between the relevant market parameters, we define now a regression analysis, which will be implemented and applied at a later point to give these relationships a real number in the context of a coefficient. The primary scope of the regression analysis is the investigation of causal relationships, or the socalled cause-and-effect relationships. In the simplest case, such a relationship can be expressed between two variables, the dependent variable Y and the independent variables X. The variables X and Y always correspond to the respective returns of a market parameter, June 27-28, 2012 Cambridge, UK 12 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 which can be determined in the simulation model. The multivariate regression approach has the following formxii : πΜπ = π0 + π1 ∗ π₯1 + π2 ∗ π₯2 + … + ππ ∗ π₯π + ππ with π: πΜπ : π0 : ππ : π₯π : ππ : (6) π·ππππππππ‘ ππππππ‘ πππππππ‘ππ πΈπ π‘ππππ‘πππ ππ π‘βπ πππππππππ‘ π£ππππππππ π − ππππππ‘ πππππππ‘ππ πΆπππ π‘πππ‘ ππππππππ π πππππ π πππ πππππππππππ‘ ππ π‘βπ ππππππ‘ πππππππ‘πππ π πΌππππππππππ‘ π£πππππππ − π ππ‘π’ππ ππ π‘βπ ππππππ‘ πππππππ‘ππ π·ππ£πππ‘πππ ππ π‘βπ ππ π‘ππππ‘ππ π£πππ’π ππ π‘βπ πππ πππ£ππ‘πππ π£πππ’π It consists of the sum of the individual influencing market parameters, together with respective regression coefficients. Due to the fact, that between the regression line and the observed values deviations exist, it can be assumed, that there is no straight line on which all the observed (x, y) – combinations belong to. The consulted market parameters are not sufficient to describe the entire process of a specific market parameter. The influencing variables which are not covered of the empirical Y- values are reflected in very low deviations from the regression line. These deviations can be represented by a vector π, whose values ππ is known as residuals. Thus, Y is additively from its systematic components, the identified influencing factors and the residual ππ . The regression coefficients ππ have an important substantive meaning, they indicate the marginal effect of the change of an independent variable on the dependent variable Y. The quantity of the regression coefficients should not be regarded as a measure of the importance of that variable. The calculation of the regression coefficients ππ is based on the minimization of the sum of squared residuals. Once all regression coefficients are calculated, it is now necessary to determine the environmental influence. In the multivariate regression model, it is not possible to integrate directly the environment, because there are no empirical variables and returns of the development of the environment. The special feature of this simulation model is exactly the June 27-28, 2012 Cambridge, UK 13 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 basis of our approach, we determine the environmental factors as well as its impact. Based on that, we imply the unexplained portion in the regression function as an influence with a regression coefficient and a variable. To realize this approach, we use the multiple determination π 2 . The multiple coefficient of determination is a global quality measure, which indicates how well the dependent variable is explained by the regression model. The basis are the calculated residuals. In order to assess these residuals, we need a benchmark. This benchmark is calculated as the difference between the observations π¦π and the mean π¦Μ . Furthermore, it requires the scattering decomposition based on the total sum of squared deviationsxiii. Thus, follows the calculation of the multiple coefficient of determination the equation: π 2 = ∑πΎ Μπ − π¦Μ )² π·πππππππ ππππππ π=1(π¦ = πΎ ∑π=1(π¦π − π¦Μ )² πππ‘ππ ππππππ with π 2 : π¦π : π¦Μπ : π¦Μ : πΎ: (7) πΆππππππππππ‘ ππ π·ππ‘πππππππ‘πππ ππππ’ππ ππ π‘βπ πππππππππ‘ π£ππππππππ π·ππ‘πππππππ ππ π‘ππππ‘ππ π£πππ’π ππ π πππ π₯π ππππ ππ’ππππ ππ πππ πππ£ππ‘ππππ (π = 1,2, . . , πΎ) The multiple coefficient of determination is a normalized value and represents values in the interval [0,1]. The greater the proportion of the explained variation to total variation is, the greater is the value of π 2 . The coefficient of determination can be determined as the square of the correlation. However, in the multivariate case the estimated variables must be formed by linear combinations of several independent variables. Therefore, π refers to a multiple correlation coefficients. By considering the environment with the multivariate regression model, we assume a bivariate context. According to this, the relationship can be expressed as: πππ₯π‘πππ,π = √1 − π 2 June 27-28, 2012 Cambridge, UK (8) 14 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 with π 2 : ππ’ππ‘ππππ πΆππππππππππ‘ ππ π·ππ‘πππππππ‘πππ πππ₯π‘πππ,π : πΆπππππππ‘πππ πππ‘π€πππ π‘βπ πππ£ππππππππ‘ πππ ππππππ‘ πππππππ‘ππ π The formula approximates the correlation of the environment with the respective multivariate regression model of each market parameter. The determined correlations are now the basis for the derivation of the influences and regression coefficients for the environment. However, one should bear in mind, that this is an approximation to the actual value. To determine the influence coefficients πππ₯π‘πππ,π , we have to adjust the calculated correlation with the standard deviation of the environment and the relevant market parameter. Thus, the formula for calculating the influence of an exogenous variable on a market parameter isxiv. πππ₯π‘πππ,π = πππ₯π‘πππ,π ∗ ππ (9) πππ₯π‘πππ with ππ : ππ‘ππππππ πππ£πππ‘πππ π‘βπ πππ‘π’ππ ππ π‘βπ πππππππ‘ππ π πππ₯π‘πππ : ππ‘ππππππ πππ£πππ‘πππ π‘βπ πππ‘π’ππ ππ π‘βπ πππ£πππππππ‘ πππ₯π‘πππ,π : πΆπππππππ‘πππ ππ π‘βπ πππ£πππππππ‘ π€ππ‘β πππππππ‘ππ π Because the environment returns are constructed as random numbers, the choice of the standard deviation is arbitrary. In this context, the standard deviation of the environment returns is set at one percent and the influence coefficients πππ₯π‘πππ,π are calculated according to equation 9. Regarding the expectation values of the normally distributed environment returns, the historical arithmetic average values of the corresponding market parameters shall be considered, to reflect the historical trend of the market. For example, the chosen economic indicator GDP, generally shows an empirical rising trend, which justifies the use of this trend. For the final determination of the expectation value π of the environment return we used the previously determined influence coefficients. According to Siebert (2010) the expectation June 27-28, 2012 Cambridge, UK 15 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 value of the environment return has to be chosen in that way, that it explains the given expected development of a market parameter, which cannot be completely described by the considered influencing factors. As soon as we simulate different scenarios, we assume an expected trend of a market parameter, and due to that the environment return will be adjusted to that trend. If we don’t consider an individual trend for a specific market parameter, we determine the expectation value of the environment return through the historical market average of the corresponding parameter and the historical averages of the influencing factors. 8 Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π ππ‘π’πππ,π‘ = πππ₯π‘πππ,π ∗ πππ₯π‘πππ π,π‘−1 + ∑ ππ ∗ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π ππ‘π’πππ,π‘−1 (10) π=1 mit πππ₯π‘πππ : ππ : Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π ππ‘π’πππ,π‘ : Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π ππ‘π’πππ,−1π‘ : πΌππππ’ππππ ππ π‘βπ πππ£ππππππππ‘ ππ π πΌππππ’ππππ ππ π‘βπ ππππππ‘ πππππππ‘ππ π ππ π πππππππππ ππ£πππππ πππ‘π’ππ ππ πππππππ‘ππ π ππ π‘ πππππππππ ππ£πππππ πππ‘π’ππ ππ πππππππ‘ππ π ππ π‘ − 1 Due to formula 10, the expectation value of the corresponding environment return can be determined through the calculated influence coefficients and the historical average values. Thus, the environment returns are given as normally distributed random numbers with the standard deviation of one percent and with the expected value πππ₯π‘πππ,π−1. The normal distributed random values are realized through the polar method of George Marsagliaxv. Hence, all the basics to calculate the development of the various market parameters are now given. The implementation results from the sum of the newly estimated regression coefficients of each market parameter plus the influence of the environment. According to this, the formula to simulate future returns of each market parameter is given by the weighted sum of the yield of individual influencing factors and the rate of change of the environment, which is simulated by generating a random number. June 27-28, 2012 Cambridge, UK 16 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 8 (11) π ππ‘π’πππ,π‘ = πππ₯π‘πππ,π ∗ π ππ‘π’ππππ₯π‘πππ π,π‘−1 + ∑ ππ ∗ π ππ‘π’πππ,π‘−1 π=1 mit πππ₯π‘πππ : ππ : π ππ‘π’πππ,π‘−1 : π ππ‘π’ππππ₯π‘πππ π,π‘−1 : πΌππππ’ππππ ππ π‘βπ πππ£ππππππππ‘ ππ π πΌππππ’ππππ ππ π‘βπ ππππππ‘ πππππππ‘ππ π ππ π π ππ‘π’ππ ππ π‘βπ ππππππ‘ πππππππ‘πππ π ππ π‘ − 1 π ππ‘π’ππ ππ π‘βπ πππ£ππππππππ‘ πππ π ππ π‘ − 1 On the basis of this multivariate regression analysis, the future price developments of the individual market parameters are simulated and ongoing the Cash Flows are calculated with the presented exposure map. Once the simulation model simulated a user-defined time period, the model returns the Cash-Flow-at-Risk to an expected value of zero or a specified benchmark. The model itself has a very flexible character to simulate the regarded world. It’s individual and company specific and can be expanded and supplemented by other relevant factors. One suggestion of the model and its application are shown and illustrated in the following chapter. EVALUATION To evaluate our simulation model, it is necessary to back test it. This is done through stress testing, to estimate how a simulated portfolio of a company responds to extreme economic conditions.. To do so, we have to follow two steps. First, we develop plausible scenarios with market fluctuations. Then the evaluation of the portfolio with respect to a given scenario follows. In this paper, the simulated portfolio includes a maximum of three different products, which are composed from the given resources of the simulation model. The three products are presented in the exposure map (table 2) and consist of four different June 27-28, 2012 Cambridge, UK 17 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 raw materials. The raw materials are aluminum, copper, nickel and zinc, which allow to produce various metal products. These raw materials should constitute the basis of this analysis. The pricing of a product is based on the partial prices of the raw materials and does not include any additional fixed costs. The market fluctuations for the stress testing are generated of the standard deviations and the corresponding expectation values of the market parameters. Moreover, due to possible variations in sales on the corporate basis, the sales of a product are controlled by its expected value and volatility. Fluctuations can occur in this context by machine breakdowns, lacks of production means or technical know-how. In addition, the company can not sell the goods in foreign countries; otherwise we have to consider additional exchange rate risks. The simulation period in all simulated scenarios is 16 quarters or four years. The choice of the four years is due to the fact, that we consider a short-term cycle in the sense of Kitchin (1923). The fluctuations in this Kitchin-cycle are caused by the storage and production of companies, which leads to a reasonable observation period. However, also exogenous effects can occur, such as wars, natural disasters or financial instability in the world's economies. The validity of a simulation is thereby strengthened and interpreted as soon as it is repeated several times. Only by this way it is possible to obtain a distribution of Cash-Flows, which can be analyzed with the confidence level. To strengthen and to increase the significance of a scenario we chose as part of the stress tests a number of 5000 simulation runs. To calculate the Cash-Flow-at-Risk to a expected value of zero, we determined for all scenarios a confidence level of 95 %. This means that with a probability of 95 % the future Cash-Flows exceed the calculated Cash-Flow-at-Risk. Once all required parameters are specified, the implementation of the various scenarios follows. Subsequently the generated Cash-Flow distributions are analyzed by using theoretical fundamentals and charts. June 27-28, 2012 Cambridge, UK 18 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Standard and 20 % Volatility: In the first two scenarios, we consider the portfolios with three different products. The layout of the products can be found in table 2. The first scenario Standard uses the obtained average returns, correlations and regression coefficients of the historical time series, to simulate future prices. In the second scenario Vola_20%, all implied volatilities of all market parameters are increased by 20 %. This test is intended to help analyzing how the model behaves with strong market fluctuations of all market parameters. The results of the simulations are shown in Figure 1. The Cash-Flow-at-Risk of scenario Standard is less than the Cash-Flow-at-Risk for the scenario Vola_20%. Furthermore, by increasing the volatility of each market parameter, the volatility of the Cash-Flow distribution is higher. This observation fulfills the expectation. The variation in the volatility of a market parameter describes the direct influence of the environment on the current market parameter. With an increased volatility, the influence of the environment is greater (Equation 9). The development of the environment is not predictable, because based on the Polar-method the future development is generated by a random walk. Therefore, the results show a possible plausible development at a higher volatility of the market parameters. The small deviation between the Cash-Flow-at-Risk results probably stems from the fact that the general development of the simulation model suggests a growing economy and thus an increase in sales. Based on the assumption that the historical time series imply a positive development of the economy and consumption, the distribution of the Standard scenario also seems to be plausible. The products and in particular the prices of the products are conceived in that way that in any case, the prices cover the variable costs and so have a positive CashFlow. Challenging economic conditions: So far we simulated a scenario based on historical time series and one scenario with an increased volatility of 20 %. However, the observation of June 27-28, 2012 Cambridge, UK 19 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 extreme market volatility is of particular interest in the evaluation of the simulation model. Therefore, we simulated in the following section, three scenarios that will test the simulation model at extreme negative developments. The first scenario BIPnull reflects a prolonged recession with no growth in the economy. The second scenario is MetalsUp which simulates the impact on the Cash-Flow by using two times the expected price increase of the raw materials. This scenario for example could arise by an increasing demand of metals from developing countries. The third scenario MetalsUp_SalesDown exacerbates the previous scenario by falling sales of all three products by 20 % caused by internal conditions. Here machine breakdowns or material defects may be responsible for a falling rate of production. The distributions of the simulated scenarios are shown in Figure 2. The Cash-Flow-at-Risk results of each scenario show, that with rising prices of raw materials, the highest probability for a negative outcome is achieved. However, the lowest mean of the Cash-Flows is achieved by the scenario with rising metal prices and dropping sales. Therefore, the average of the lowest Cash-Flow is generated by the third scenario. Moreover, the volatility of the third scenario is significantly lower and the concentration of the results to a particular sample space is higher. This concentration results from the simultaneous decrease in the rate of production and the rising prices. On the one hand, the sales decrease and on the other hand, the costs increase significantly. The results of the three scenarios seem also to be plausible. The simulated Cash-Flows with the increased prices for raw materials and the stagnant economy are significantly lower than the previous scenarios. On the one hand, the production of products gets more expensive, and thus reduces the profit margin significantly. On the other hand, through a stagnant economy the sales are hampering and the revenue is collapsing. June 27-28, 2012 Cambridge, UK 20 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Booms and economically beneficial scenarios: After we examined negative developments, which can occur, we are now analyzing economically beneficial scenarios. There are two different scenarios constructed. These scenarios are based on the exact opposite trends as the previously considered scenarios. The first scenario implies a recovering economy as it occurred in the years 2009 and 2010 in Germany. The Federal Statistical Office of Germany recorded in the 3rd Quarter of 2009 a growth of the gross domestic product of 2.3 %. Therefore, in the first scenario we assume a growth in gross domestic product of 2.5 %, which should contribute to increasing sales. The second scenario implies falling metal prices, which should lead to lower production costs. The prices of raw materials develop exactly opposite to the average empirical development of the last decades. The distributions of simulated Cash-Flows are shown in Figure 3. As you can see in the two charts, an economic upturn has a greater positive influence on the market and the production volume and therefore on the Cash-Flow development, as falling prices. Falling prices increase the profit margin but not the sales and thus the revenue. Comparison of considered business cycles: Finally, we are now comparing the different results from three different types of scenarios. The goal is to contrast the different distributions of Cash-Flows. Figure 4 shows the simulated Cash-Flows of the scenarios Standard, BIPnull and BIPup. In the diagram you can see clearly the differences between the two extreme scenarios and the standard scenario. The shifts in the Cash-Flow distributions show that the simulation model is able to generate possible future Cash-Flow developments, based on the changed market parameters. Figure 5 shows the shift of the Cash-Flows with falling and rising commodity prices compared to the simulation with the average empirical development based on historical time series. However, it can be recognized that the shifts of the distributions to the left and right on the X-axis are lower as in the previous scenarios. This June 27-28, 2012 Cambridge, UK 21 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 means that the impact of falling and rising commodity prices has a smaller impact on the future Cash-Flow development in comparison to cyclical market fluctuations. Figure 6 illustrates the Cash-Flow-at-Risk results of each scenario. The diagram shows that the results are similar for the same respective circumstances. With an increasing economic activity and falling prices of raw materials, a significantly higher Cash-Flow-atRisk can be determined, in comparison to the standard development and in particular to rising commodity prices and the sluggish economy. Furthermore, the difference between an increase in volatility and the simulation with the average historical development is evident. The Cash-Flow-at-Risk is lower, equally the mean of the scenario Vola_20%. However, the chart 1 shows that the distributions are very similar. In scenario Vola_20% you would expect to see the consequences of the higher deviations with a more significant effect in results in comparison with the Standard scenario. This is not observable, because the model implicitly takes coherences between market factors into consideration and that leads to the resulting distribution. The Cash-Flow-at-Risk result for the scenario MetalsUp is almost zero, that’s why it cannot be seen in the diagram 6. The result is mapped between the result of the scenarios BIPnull and MetalsUp_SalesDown. In Table 3 are all the results of the Cash-Flow-at-Risk calculations of each scenario presented. Like mentioned before, the simulated Cash-Flows aren’t including any fixed costs the company faces through the production of the products. Through that assumption, the Cash-Flow-at-Risk results in the following table seem big and aren’t negative or zero for the scenarios you would expect that. However, the Cash-Flow-at-Risk results of the different scenarios differ significantly and show that the simulation model works as it was expected. In summary, we conclude that the simulation model explains the relationships of the various market parameters well. The generated simulations show that the expected results occur, and thus the robustness and stability of the model is proved. Even in extreme market June 27-28, 2012 Cambridge, UK 22 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 fluctuations, the simulation model provides reasonable results and does not reject the calculated correlations. CONCLUSION The simulation model is based on System Dynamics, bivariate and multivariate regression analysis and also on the principle of random walk and the Cash-Flow-at-Risk approach. This macroeconomic market model is able to simulate and aggregate in the context of a corporate exposure map and the inclusion of business-relevant market factors, future Cash-Flows. The simulation model is capable of simulating different scenarios with different configurations of the various market parameters. The scenario analysis shows that the model has the required quality and robustness to deliver plausible results even in extreme market situations. Moreover, an adaptation of the implementation on other companies is feasible with relatively little effort, based on the circumstance of the individual configuration. The fact that the model uses the multivariate regression analysis to measure the direct influence of individual market parameters on other factors increases the validity and reflects the theoretical and the empirical correlations in more detail. The consideration of an exogenous variable which is defined as the environment and the evaluation of an influence of that variable on the development of the considered market parameters is in contrast to a lot of other models. However, not only the implication of an exogenous variable, also the consideration of historical returns to predict the future development, differs from various models. The use of empirical data is efficient and often the only possibility for quantifying theoretical relationships. However, the projection of empirical relationships for the future can lead to incorrect results. Historical data deliver just snapshots June 27-28, 2012 Cambridge, UK 23 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 of possible developments, which will not recur in the same constellation in the future. That’s why it is possible that through structural breaks in the context of the market dynamic these empirical correlations lose their validity. The problem in this context is the low frequency of data to describe stable relationships. Starting points for future studies would be in the observation of the nonlinearity of the regression coefficients. Bartram (1999) describes the non-linearity as the complexity of the exposure and the difficulty of a correct evaluation of financial risks through the capital markets. That could lead to small changes in exchange rates, which are superimposed of other price relevant information. This means that financial market participants consider only large exchange rate fluctuations in the business valuation or Cash-Flow calculation. In addition, the use of historical data to forecast future prices is in the literature discussed with controversy. Above that, the model shows a flexible character with the possibility of implementing further enhancements. That could be considering default risk within supply or storage costs due to fluctuation in demand. June 27-28, 2012 Cambridge, UK 24 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 TABLES t-1\t GDP GDP X CPI X EUR/USD X Interest Alu. Copper Nickel Zinc CPI X X EUR/USD Interest X X X X X X X X X X X X X X X Alu. X X X X Copper X X X X Nickel X X X X Zinc X X X X Table 1: Matrix of the influencing coefficients Product 1 Income elasticity Price elasticity Price Sales Sales μ Sales σ Aluminium Copper Nickel Zinc 1 -1 1000 3000 0 0 0.2 0.02 0.005 0.005 Product 2 Product 3 0.5 1.2 -1 -1.5 1100 40 3000 15000 0 0 0 0 0.3 0.01 0.02 0.001 0.001 0 0 0 Table 2: Stresstesting Exposure Map CFaR Mean of the CF‘s Standard deviation Standard 30,372,894.69 56,600,668.43 14,475,630.82 Vola_20% 23,184,192.12 56,283,574.33 17,779,785.99 BIPnull 2,417,715.73 35,492,772.54 18,699,541.03 MetalsUp 15,330.14 35,717,172.65 19,902,825.30 MetalsUp_SalesDown 5,068,065.91 17,718,321.23 7,124,585.93 BIPup 87,656,579.66 100,611,857.40 7,234,513.07 MetalsDown 63,029,150.07 79,346,077.07 9,010,692.07 Table 3: Results of the scenarios June 27-28, 2012 Cambridge, UK 25 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 FIGURES Figure 1: Standard development and volatility increases by 20% Figure 2: Sluggish economy, rising commodity prices and sales decline June 27-28, 2012 Cambridge, UK 26 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Figure 3: Improving economy and falling commodity prices Figure 4: Comparison of standard development with a stagnating and attractive economy June 27-28, 2012 Cambridge, UK 27 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Figure 5: Comparison of standard development with falling and rising commodity prices Figure 6: Cash-Flow-at-Risk distribution of scenarios June 27-28, 2012 Cambridge, UK 28 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 REFERENCES Backhaus, K., & Erichson, B., & Plinke, W., & Weiber, R. (2008). Multivariate Analysemethoden – Eine anwendungsorientierte Einführung. Berlin: Springer – Verlag Bartram, S. M. (1999). Corporate Risk Management – Eine empirische Analyse der finanzwirtschaftlichen Exposures deutscher Industrie- und Handelsunternehmen. Bad Soden/Ts: Uhlenbruch Duch, J. (2006). Risikoberichterstattung mit Cash-Flow at Risk-Modellen. Frankfurt am Main: Lang Junius, K., & Kater, U., & Meier, C. P., & Müller, H. (2002). Handbuch Europäische Zentralbank : Beobachtung, Analyse, Prognose. Bad Soden/Ts.: Uhlenbruch Kim, J., & Malz, A., & Mina, J. (1999). LongRun Technical Document. New York: RiskMetrics Group Kitchin, J. (1923). Cycles and Trends in Economic Factors. The Review of Economics and Statistics, 5(1), 10–16 Lang, C. (2005). Theoretische und empirische Aspekte der Prognose wichtiger makroökonomischer Größen. Göttingen: Cuvillier Lee, A. (1999). CorporateMetrics – The Benchmark for Corporate RiskManagement. New York: RiskMetrics Group Rothengatter, W., & Schaffer, A. (2008): Makro kompakt – Grundzüge der Makroökonomik. Heidelberg: Physica – Verlag Seese, D., & Siebert, L., & Vogel, A. (2011). Risikomanagement durch Modellierung eines makroökonomischen Marktmodells im Kontext unternehmensweiter Stresstests. Risikomanager 15/2011 Seese, D., & Schlottmann, F., & Vogel, A. (2011). Market modelling for anticipating risk in a context of macroeconomic stresstest. Proceedings of International Business Research Conference. Dubai Siebert, L. (2010). Modellierung eines makroökonomischen Modells: Bachelor Thesis, Karlsruhe Institut für Technologie (KIT), Karlsruhe, Germany Spremann, K., & Gantenbein, P. (2007). Zinsen, Anleihen, Kredite (4th ed.). München: Oldenbourg Stein, J., & Usher, S. E., & LaGatutta, D., & Youngen, J. (2001). A Comparables Approach To Measuring Cash-Flow-at-Risk for Non-Financial Firms. Journal of Applied Corporate Finance, 13(4) June 27-28, 2012 Cambridge, UK 29 2012 Cambridge Business & Economics Conference ISBN : 9780974211428 Thome, H. (2005). Zeitreihenanalyse: eine Einführung für Sozialwissenschaftler und Historiker. München: Oldenbourg Wissenschaftsverlag Turner, C. (1996). VaR as an industrial tool. Risk, 9(3) Veith, J. (2006). Bewertung von Optionen unter der Coherent Market Hypothesis. Wiesbaden:Deutscher Universitats-Verlag - GWV Fachverlage GmbH i The developed model is an enhancement. For further details see also Seese, Siebert, Vogel (2011) or Seese, Schlottmann, Vogel (2011) ii Income and price elasticity need to be estimated by internal research processes. They are individual and depend on the company’s products. iii See Veith (2006) iv Those parameters turned out to be the most relevant market drivers of the company having a closer look at for sample reasons v See http://www.destatis.de vi See Rothengatter, Schaffner (2008): vii See http://www.bundesbank.de viii See Junius, Kater, Meier, Müller (2002) ix See Spremann, Gantenbein (2007) x See http://www.destatis.de/jetspeed/portal/cms/Sites/destatis/Internet/DE/Presse/abisz/VPI, templateId=renderPrint.psml xi See Lang (2005) xii See Backhaus, Erichson, Plinke, Weiber (2008) xiii See Backhaus, Erichson, Plinke, Weiber (2008) xiv See Thome (2005) xv See Veith (2006) June 27-28, 2012 Cambridge, UK 30